Sources and connectors
A source is a system an agent works with: a database, a reporting environment, an ETL tool or a document store. Each source belongs to one customer, has a kind and an environment (production or test, for example), and its own data access setting.
Databases (native)
For these databases the platform has its own fixed query templates. An agent gets metadata, counts and statistics without free SQL being sent to the source.
| Database | What agents can do |
|---|---|
| SQL Server / Azure SQL | Metadata, quality, performance, ETL log profiles, verification sandbox, free SQL if the source is released for it |
| PostgreSQL | Metadata, quality and part of the performance tools |
| SAP HANA Cloud | Metadata and quality |
Connectors
Connectors link systems that are not databases. They run through the platform: the agent never receives a password or token for such a system itself.
| Category | Connectors |
|---|---|
| Reporting | Power BI, SAP Analytics Cloud |
| Data platform | Microsoft Fabric, Databricks, SAP Datasphere |
| ETL and transformation | Azure Data Factory, dbt Cloud, Matillion |
| Documents | SharePoint, OneDrive, Google Drive |
| Development | Azure DevOps, GitLab |
| Your own | Any MCP server of the customer; see your own MCP servers |
With Power BI an agent can read semantic models and edit definitions. Whether a connector may write is set per source. A write action always goes through a proposal that a person approves.
Adding a source
- Choose the customer. A source always belongs to one customer.
- Kind and environment. For example SQL Server, environment production.
- Connection. Server, database, login. The password is stored encrypted per source. After that the app shows only a fingerprint and the moment it was set.
- Connection test. The platform tries to connect and tells you what goes wrong.
- Data access. You choose the setting: metrics only, encrypted values or open. See data access.
- Label. A short name that agents get to see. Never put the customer's name in it.
Per source an agent sees only an id, the kind, the environment and the label. It does not get the technical name or the key of the source.
Sources per work item
A person decides which sources belong to a work item. Within that set the agent chooses what to work on. A subtask cannot name a source outside the set. An agent answering a finance question therefore does not end up on the HR database by accident.
Changing or disabling a source
You can change the connection, the label and the setting. Not the source's key: it is recorded in the ground rules, and a rename would silently move permissions. If you no longer want to use a source, disable it. Deleting is only possible once no work item, dataset or ground-rules version refers to it.
Example
A retailer connects three sources: the Azure SQL warehouse (production, setting encrypted values), the Power BI workspace with the sales model (read only) and the Azure DevOps repository with the ETL code. A work item "why does revenue in the report differ from the warehouse?" gets all three. The agent compares the DAX definition with the SQL view, finds a filter on returns that exists only in the model, and proposes a change in the repository. A person approves it.